|
The inherently spatial uncertainty of many geographic objects has been the target of several, though still limited number of, studies. Two types of uncertainties, namely fuzziness and randomness, have been formally characterized within the framework of fuzzy set and probability theories, respectively. However, the scale issue has not been explicitly considered in modelling vagueness, whilst it is true that the degree of uncertainty of many objects is in relation to the scale of its representation. Furthermore, though fuzzy data types have been there for some years, a computational framework for handling fuzzy spatial objects is still lacking. In this article a previously introduced neural representation of polygon layers has been generalized to represent not only polygons but also points, lines, and complex combinations of these in the hard (crisp) and fuzzy domains. Two types of neural units, combined with two types of activation function, were identified as the processing core of the model, where the activation function can be either hard or fuzzy. In the hard case, we show how it is possible to differentiate among interior, exterior and boundary of a polygon by using a tri-valued activation function instead of the binary function originally used. The generalization to fuzzy domains can be implemented in computers, allows hierarchical constructions of complex objects from basic ones and allows us to build upon traditional spatial databases (with crisp boundaries). It is shown how the degree of fuzziness may be related to the scale of the representation under the premise that a decrement in the degree of fuzziness may lead to new details becoming apparent in the boundary. Indications on how to perform a complex overlay of fuzzy maps are also provided.
|
